lable at ScienceDirect

Tourism Management 46 (2015) 222e232

Contents lists avai

Tourism Management

journal homepage: www.elsevier .com/locate/ tourman

Exploring the travel behaviors of inbound tourists to Hong Kong usinggeotagged photos

Huy Quan Vu a, 1, Gang Li a, 2, Rob Law b, 3, Ben Haobin Ye c, *

a School of Information Technology, Deakin University, Vic 3125, Australiab School of Hotel & Tourism Management, The Hong Kong Polytechnic University, Hong Kongc School of Business, Sun Yat-sen University, No. 135, Xin Gang Xi Road, Guangzhou, China

h i g h l i g h t s

� A novel approach to travel behavior analysis is introduced, based on geotagged photos.� A dataset comprises photos on Flickr by Hong Kong inbound tourists is built.� The spatial and temporal information of photos infers tourists' movement trajectories.� The advantage of the approach is shown by analyzing inbound tourist travel behavior.� The study benefits destination development, transportation planning, and impact management.

a r t i c l e i n f o

Article history:Received 3 November 2013Accepted 2 July 2014Available online

Keywords:Data miningGeotagged photoTravel behaviorGlobal positioning system

* Corresponding author. Tel.: þ86 84114091; fax: þE-mail addresses: hqvu@deakin.edu.au (H.Q. Vu),

hmroblaw@polyu.edu.hk (R. Law), yehb3@mail.sysucom (B.H. Ye).

1 Tel.: þ61 432 411 359.2 Tel.: þ61 3 9251 7434; fax: þ61 3 9251 7604.3 Tel.: þ852 3400 2181; fax: þ852 2362 9362.

http://dx.doi.org/10.1016/j.tourman.2014.07.0030261-5177/© 2014 Elsevier Ltd. All rights reserved.

a b s t r a c t

Insight into tourist travel behaviors is crucial for managers engaged in strategic planning and decisionmaking to create a sustainable tourism industry. However, they continue to face significant challenges infully capturing and understanding the behavior of international tourists. The challenges are primarilydue to the inefficient data collection approaches currently in use. In this paper, we present a newapproach to this task by exploiting the socially generated and user-contributed geotagged photos nowmade publicly available on the Internet. Our case study focuses on Hong Kong inbound tourism using29,443 photos collected from 2100 tourists. We demonstrate how a dataset constructed from suchgeotagged photos can help address such challenges as well as provide useful practical implications fordestination development, transportation planning, and impact management. This study has the potentialto benefit tourism researchers worldwide from better understanding travel behavior and developingsustainable tourism industries.

© 2014 Elsevier Ltd. All rights reserved.

1. Introduction

The tourism industry plays a major role in economic develop-ment for many countries and regions, helping to improve thelivelihoods of residents (Ashley, Brine, Lehr, & Wilde, 2007). Bothpublic and private sector tourism operations are highly dependenton sustainable development planning, which aims to createappropriate employment, maintain the natural environment, and

86 84036924.gang.li@deakin.edu.au (G. Li),.edu.cn, damonyhb@hotmail.

deliver a high-quality visitor experience. Due to the perishablenature of the tourism industry, an accurate understanding of travelbehaviors is crucial (Edwards, Griffin, Hayllar, Dickson, &Schweinsberg, 2009).

Tourism managers have long been seeking insights into travelbehavior for the purposes of destination management, productdevelopment, and attraction marketing (Li, Meng, & Uysal, 2008).For instance, good transportation planning enables tourism pro-viders to meet tourists' needs and coordinate their travel with localtransportation flows. However, it requires planners to know abouttourist preferences, daily itineraries, and the factors that influencethem. Movement information can also be used to identify bottle-necks and unnecessary barriers in the flow between accommoda-tion and attractions, or any other tourist destinations (Prideaux,2000). In tourism location development, knowledge about trav-elers' location preferences can be used to redefine existing

H.Q. Vu et al. / Tourism Management 46 (2015) 222e232 223

attractions, plan new ones, and market both more effectively (Lew&McKercher, 2006). The identification of the actual routes taken bytourists during their trips can help define the boundaries of dis-tricts and nodes, as well as the most appropriate gateways. Thisinformation can be used to develop new attractions and productsalong common routes as well as in district and destination nodes(Chancellor, 2012). In terms of impact management, it is importantfor tourism managers to identify the time and space characteristicsof the routes and destinations which tourists visit most frequentlyin order to develop appropriate plans to prevent capacity overload,which has the potential for negative social, environmental, or cul-tural impacts (Lew & McKercher, 2006).

Some prior research focuses on developing techniques foranalyzing tourist travel behaviors. For instance, Xia, Spilsbury,Ciesielski, Arrowsmith, and Wright (2010) introduce a method fortourist market segmentation based on dominant movement pat-terns in a case studyof Phillip Island, Australia (Xia et al., 2010). Dengand Athanasopoulos (2011) utilize an anisotropic dynamic spatiallag panel OrigineDestination (OD) travel flowmodel to understandAustralian domestic and international inbound travel patterns.Leung et al. (2012) use content and social network analyses toexamine online trip diaries and map overseas tourist movementpatterns during the Beijing Olympics. Such growing academicattention being paid to travel behavior indicates that this is aninteresting topic which is important in the planning and decision-making processes of tourism managers.

Despite these efforts, both researchers and managers still find itdifficult to fully capture and understand the travel behavior of inter-national travelers. The difficulties are due to the following barriers:

Information Capture: Survey and opinion polls are popularmethods to collect travel data from tourists (Asakura & Iryo,2007; Lew & McKercher, 2006; McKercher & Lau, 2008). How-ever, these approaches are usually time consuming and limitedin terms of the number of responses as well as the scale of theinformation captured. The data are normally unable to reflectactual travel patterns closely (Zheng, Zha, & Chua, 2011). Thereis considerable need for a more efficient method of capturingcomprehensive travel behavior data from tourists.Travel Preferences: It is a natural assumption that travelbehavior will vary among different groups of tourists (Batra,2009). For instance, tourists from different countries may havedifferent preferences for their length of stay and the attractionsthey want to visit, and this will, in turn, affect their travel ac-tivities (Leung et al., 2012). There have been limited attempts toanalyze travel behaviors in a way which takes these preferencesinto consideration.

Recently, advances in multimedia and mobile technologies haveallowed large volumes of user-generated data, such as travelphotos, to be created and shared. Many photo-capturing devices,like smartphones and tablets, now have built-in global positioningsystems (GPS) technology which enable geographical information(latitude and longitude coordinates) to be stored as metadata witheach photo a user takes. Those geotagged photos, with time andgeographical information embedded, allow the spatialetemporalmovement trajectories of the user to be inferred. One of the mostpopular online resources for people to share their travel experi-ences by uploading photos is Flickr (www.flickr.com). Containingmillions of geotagged photos, Flickr is a rich data source for miningtourist travel patterns (Lee, Cai, & Lee, 2013; Zheng, Zha, & Chua,2012). Research into the nature and applications of georeferencedmultimedia is an emerging topic in Computer Science (Zheng et al.,2011), with many attempts having recently been made to developtools and techniques for data mining (Kalogerakis, Vesselova, Hays,

Efros, & Hertzmann, 2009; Yanai, Kawakubo, & Qiu, 2009; Yanai,Yaegashi, & Qiu, 2009). It will, therefore, be advantageous fortourism researchers to adopt these advanced technological de-velopments when studying travel behavior.

In this paper, we attempt to address the shortcomings in theexisting literature on tourist travel behavior by utilizing the geo-tagged photos that are available on social networking sites. Firstly,we present a method for constructing the data collection processwhich captures travel information from geotagged photos on Flickr.Then, we describe two relatively new data-mining techniques forprocessing and analyzing these data to yield information about thetravel behaviors of tourists. This case study focuses on inboundtourism inHongKong, amajor tourismdestination in theAsia Pacificregion. The study aims to discover the locations inwhich tourists aremost interested and the routes they take when visiting Hong Kong.This method and the associated findings are of potential benefits totourism researchers worldwide who are interested in travelbehavior, and will also help tourism managers in Hong Kong andlikely elsewhere to construct plans for sustainable development.

Having thus set the context for undertaking this work, the rest ofthe paper is organized as follows. In Section 2, we review existingwork, which uses global positioning technology such as GeographicInformation Systems (GIS) to analyze travel behavior. We thenprovide a summary of the techniques developed for processinggeotagged photos in travel analysis, and define our research ob-jectives. Section 3 presents the methods for extracting andanalyzing geotagged photos. We demonstrate their effectiveness ina case study presented in Section 4. Finally, Section 5 concludes ourpaper and offers future research directions.

2. Related work

2.1. Travel behavior analysis using geographical information

Since GIS was firstly introduced, many studies have used it toexplore the movement patterns of tourists (Lau & McKercher,2006). For instance, Wu and Carson (2008) use GIS to identifymultiple destination travel behavior for travelers in South Australia.McKercher and Lau (2008) apply it to examine the movements oftourists within an urban destination in Hong Kong, and identify 78discrete movement patterns. Other researchers use GPS to exploretourists' experiences and mobility (Zakrisson & Zillinger, 2012), orto chart visitor movement patterns in natural recreational areas(Orellana, Bregt, Ligtenberg, & Wachowicz, 2012). Additionally,McKercher, Shoval, Ng, and Birenboim (2011) compare and contrasttravel behaviors between first-time and repeat visitors in HongKong using both GPS and GIS.

Attempts have also been made to study tourist travel behaviorsusing these methods. For instance, Hwang, Gretzel, and Fesenmaier(2006) examine international tourists' multi-city trip patternswithin the United States. A large-scale study of the spatial pattern oftouristflows among the selected Asia-Pacific countries over a 10-yearperiod is presented by Li et al. (2008). Asakura and Iryo (2007) look attourist travel behavior inKobe, Japanusing trackingdata collected viamobile instruments. Smallwood, Beckley, and Moore (2012) explorethe movement patterns of visitors traveling within protected areasusing various modes of travel through a case study of the NingalooMarine Park, in north-western Australia. Masiero and Zoltan (2013)analyze the factors influencing both the spatial extent of the desti-nation visited and the selected transport mode.

2.2. Geotagged photos for travel analysis

Advances in information technology, especially the introductionof GPS technology and mobile photo capturing devices, now allow

H.Q. Vu et al. / Tourism Management 46 (2015) 222e232224

for geographical information to be available and stored in a phototag. Computer Science researchers have therefore started todevelop techniques for using these geotagged photos given thatthey are now made publicly available on many photo-sharing sitessuch as Flickr and Paranomia (Kennedy, Naaman, Ahern, Nair, &Rattenbury, 2007). For instance, Jaffe, Naaman, Tassa, and Davis(2006) attempt to develop a technique that automatically sum-marizes and visualizes a large collection of georeferenced photo-graphs. Also, Snavely, Seitz, and Szeliski (2008) propose anapproach called Photo Tourism which enables the three-dimensional (3D) modeling and reconstruction of world sitesfrom user-generated photos available on Google Maps. Quack,Leibe, and Gool (2008) mine events from community photo col-lections on a worldwide scale using geotagged photos on Flickr.Hays and Efros (2008) propose a method for automatically esti-mating geographical location for photos without geotags, a tech-nique which has the potential to generate even more data forgeographic-related research. Kalogerakis et al. (2009) introduceanother way to infer geographic location using sequences of time-stamped photos.

Different attempts have been devoted to develop intelligenttourism recommendation system using geotagged photos. Namely,Kurashima, Iwata, Irie, and Fujimura (2010) propose a probabilisticbehavior model for recommending travel path based on the pasttravel information of tourists on Flickr. Okuyama and Yanai (2013)make use of actual travel paths extracted from a large number ofonline geotagged photos to develop a travel planning system.Majid and colleagues propose methods for tourism locationrecommendation that are relevant to user preference and travelcontext (Majid, Chen, Chen, Mirza, & Hussain, 2012; Majid et al.,2013). Shi, Serdyukov, Hanjalic, and Larson (2013) propose amethod for personalized landmark recommendation. Yin, Wang,Yu, and Zhang (2012) facilitate tourist trip planning by a travelpath search system using geotagged photos. Similarly, Li (2013)proposes a travel planning system to design multi-day and multi-stay travel plans.

Due to the demand for knowledge about travel behavior, somemethods have also been proposed to mine travel patterns fromgeotaggedphotos.Mamei, Rosi, andZambonelli (2010)put forwardatechnique for the automatic analysis of geotagged photos for Intel-ligent Tourist Services. Rugna, Chareyron, and Branchet (2012) showthat the country of origin of photo owners can be identified based onthe photos they post online. Other researchers have mined tourists'travel patterns such as traffic flow between different locationswithin a destination (Zhenget al., 2011, 2012). Lastly, Lee et al. (2013)use clustering and association rule techniques to identify touristattractions and their associated movement patterns.

2.3. Research objectives

Understanding travel behavior is important for tourism practi-tioners seeking to predict the market or to provide suitable triprecommendations. The behavior pattern discovery process can besupported by different types of data, which have traditionally beencollected using surveys, questionnaires, and opinion polls. Theseapproaches are usually time consuming and unable to accuratelyreflect the travel behavior of tourists. Even with the support oftechnology such as GIS, participants are usually required to carrymobile devices for recording their movements while traveling. Thedata collected are, therefore, limited in terms of the number ofresponses or the scale of the geographical areas included.When thecollected data are not reliable or adequate, it is necessary to resortto data that are available online. The development of ComputerScience research on geotagged photos presents a new direction forcapturing tourist movement data in an efficient and timely manner.

Withmillions of geotagged and time-stamped photos available on aglobal scale, photo-sharing sites are potential gold mines for re-searchers seeking to extract data and study travel behavior.

Although prior efforts have been made in utilizing geotaggedphotos to support travel planning and recommendation of tourists,limited work focused on analyzing travel preference of tourists tosupport for strategic planning and decision making for tourismmanagers. Some examples of potential questions ask by managersare “What are the preferred locations for each group of tourists whenvisiting a tourist destination?”, “When do they prefer to visit suchlocations?” and “What travel routes are they likely to take whenvisiting different locations?”. Unfortunately, none of the existingworks attempted to discover such insightful knowledge. As such, itis still a challenging task for tourism managers to develop effectivedestination management and transportation plans to accommo-date the increasing demand of international tourists.

This paper aims to address such challenges by exploiting thesocially generated and user-contributed geotagged photos that arepublicly available on the Internet. Our specific objectives can bedefined as follows:

� to introduce a framework for effectively extracting geographicalinformation from geotagged photos posted online and using thisto analyze tourist travel behavior;

� to identify the attractions of interest to tourists with differentprofiles who are visiting a tourist destination such as HongKong; and

� to identify the travel behaviors of tourists, travel route and traveltime, in order to support traffic management and the productdevelopment of tourism businesses.

3. Methodology

In this section we firstly outline the data-collection processinvolved in extracting geotagged photos from Flickr. Then, wediscuss a number of specific challenges which need to be addressedin order to analyze this kind of data. While the massive volume ofshared photos available on Flickr is a comprehensive resource forstudying travel behaviors, the data can be noisy or misleading,especially when many photos have been taken in transit ratherthan at the attractions themselves. In addition, the sequence ofphotos, in some situations where tourists move from one attractionto another, implies mobility information, which is a key to inferringpersonal activities, intuitions, and goals. However, photos are staticmedia, while travel behaviors are dynamic. The photos shouldtherefore be transformed into a suitable representation for thetravel analysis task. Therefore, we introduce two relatively newdata mining techniques, based on density clustering and the Mar-kov Chain, to tackle these issues.

3.1. Geographic data extraction from geotagged photos

Geotagged photos are available for public view through webapplications such as Flickr, but are not directly downloadable. Theymust be accessed via Flickr's Application Programming Interface(API), documentation for which is available at www.flickr.com/services/api. One of the challenges in extracting data from thissource is that it is impossible to identify individual owners whosephotos should be downloaded. Therefore, we propose to addressthis task by firstly searching for geotagged photos taken in theenvirons of Hong Kong and then extracting the owners' informa-tion in order to retrieve the photo information.

We define a bounding box for the region fromwhich wewant toextract geotagged photos. Let xmin, ymin, xmax, and ymax represent itsgeographical coordinates for the minimum longitude, minimum

H.Q. Vu et al. / Tourism Management 46 (2015) 222e232 225

latitude, maximum longitude, and maximum latitude, respectively.A set of M seed photos are randomly extracted using Flickr's photosearch function with the bounding box. Those photos are returnedfrom the search engine together with their owner's identificationnumber (ownerID). It is possible for multiple photos to be uploadedby the same owner. We then construct a list of owners ⟨o1, o2, …,on⟩, who have uploaded photos taken in the target area. TheirownerIDs are used to retrieve the user demographic informationsuch as country of origin using the Flickr user search function. Thishelps us to differentiate Hong Kong residents from internationaltravelers. Based on the ownerIDs of the targeted tourist group andthe bounding box coordinates, we retrieve the entire collection oftheir shared photos to ensure that their travel activities arecaptured completely. It should be noted that in addition to thespatial information, each photo is usually stamped with the timeand date on which it was taken. This temporal information is alsouseful in constructing a movement trajectory for travel patternanalysis. The Flickr photo search function also allows the user tospecify temporal information to limit the search. If values for theminimum (tmin) and maximum time (tmax) are provided, onlyphotos taken between that period are returned.

3.2. Density clustering for tourism attraction identification

Geographical location is one of the most important memorycues for recalling previous trips. Many photos taken by the touristmay be related to a single day's activity, and some may also betaken on the journey there. According to Zheng et al. (2012), atourism attraction is defined as a spatial extent within ageographical location through which considerable volumes oftourist movement trajectories pass, or where many tourists visitand take photographs. Previous work proposes a Density-basedspatial clustering of applications with noise (DBSCAN) (Ester,Kriegel, Sander, & Xu, 1996), to perform this task (Kisilevich,Keim, & Rokach, 2010; Lee et al., 2013; Zheng et al., 2012). How-ever, a shortcoming of this technique is that generic photo pointsare treated as being of equivalent importance, whereas the ownersof photos are the main factor determining the importance of acluster in the geotagged data analysis context. Thus, we use a newand more advanced version of DBSCAN, namely P-DBSCAN(Kisilevich, Mansmann, & Keim, 2010), to perform this task. It wasspecifically developed for clustering geotagged photos by takinginto account information about the photo owners in this compu-tation. The details of this technique are described below.

Suppose D is a collection of geotagged photos. Each photo is apoint p referenced by a value pair ⟨xp, yp⟩ for longitude and latitudecoordinates. The distance between two photo points p and q isdenoted by Dist(p, q). The neighborhood of a photo point p, denotedby Nd(p), is defined by:

NdðpÞ ¼ ðq2D;OwnerðqÞsOwnerðpÞjDistðp; qÞ � dÞ (3.1)

whereOwner(q)¼ (oi2O) is an ownership function. In otherwords,a photo q is in the neighborhood of another photo p if they belong todifferent users and the location of photo q is within a neighborhoodradius d fromphotop. LetNeighborOwner(p) be the owner numberof

P ¼

0BB@

0 Pða2ðnþ 1Þja1ðnÞÞ … Pðamðnþ 1Þja1ðPða1ðnþ 1Þja2ðnÞÞ 0 … Pðamðnþ 1Þja2ð« « 1 «Pða1ðnþ 1ÞjamðnÞÞ Pða2ðnþ 1ÞjamðnÞÞ … 0

the neighbor photos Nd(p), and l be the owner number threshold. Aphoto p is called a core photo if its neighbor photos belong to at leasta minimum number of owners (NeighborOwner(p) � l).

The clustering process of P-DBSCAN starts with a set of unpro-cessed photos P ¼ {p1, p2, …}. For each photo pi, if it is not a corephoto it is marked as noise. Otherwise, it is assigned to a cluster cj,and all of its neighbors Nd(pi) are put into a queue for furtherprocessing. Each photo pij 2 Nd(pi) is then processed and assignedto the current cluster cj until the queue is empty. This process re-peats for the rest of the unprocessed photos in P. After obtainingclusters of photos, we examine their geographical coordinates todetermine the name and spatial extent of the tourist attractionscaptured, which we define as the Area of Interest (AOI). Here, thevalues of d and l are determined based on the scale of specificapplications. If the region under study is at themacro level such as acountry, then the AOI may be as big as a city. Large values can beassigned to d and l. If the region is at a micro level such as a park,AOI may be on a much smaller scale, and d and l take smallervalues. The implementation of this technique is discussed further inour case study in Section 4.2.1.

3.3. Using the Markov Chain for travel pattern mining

Tourism managers are interested in not only where touriststravel, but also how they get there. This section describes a method,based on the Markov Chain, for mining such travel patterns androutes taken by tourists between main attractions (Xia,Zeephongsekul, & Arrowsmith, 2009).

Let A ¼ {a1, a2, …, am} as denote an AOI as identified by P-DBSCAN. The travel trajectory of a tourist from time t ¼ 1 to t ¼ k isdefined as T ¼ {ait1, ait2,…, aitk} and the probability of tourist movingto a location ai is computed by:

P�aani

��atn�1i ; atn�2

i ;…; atoi

�¼ P

�atni

���atn�1i

�(3.2)

Equation (3.2) implies the conditional independent assumptionMarkov Chain, which can be used to model how tourists flow fromone location to another. The transition probability of a touristmoving from attraction ai to attraction aj(is j), between time t ¼ nand t ¼ n þ 1, is computed by:

P�ajðnþ 1ÞjajðnÞ

� ¼ P�ajðnþ 1Þ∩aiðnÞ

�PðaiðnÞÞ

(3.3)

where the numerator is the probability of the tourist visiting bothlocations ai and aj, with ai being visited first followed by aj. Theevent ai(n) can be expressed as a combination of the mutuallyexclusive events ∪k

j¼1ðaiðnÞ∩ajðnþ 1ÞÞ, thus the denominator iscomputed by:

P�ajðnÞ

� ¼Xm

j¼1

P�aiðnÞ∩ajðnþ 1Þ� (3.4)

The transition probability for all possible travel routes amongdifferent attractions can be computed and presented in a one-steptransition probability matrix P:

nÞÞnÞÞ

1CCA

Table 2Data collection.

Group Number of tourists Number of photos

Western tourist 1036 Tourists 15,990 PhotosAsian tourist 1064 Tourists 13,453 PhotosTotal 2100 Tourists 29,443 Photos

H.Q. Vu et al. / Tourism Management 46 (2015) 222e232226

where the value of each entry Pij 2 P reflects how likely tourists areto travel from attraction ai to attraction aj. A demonstration of thistechnique for travel path analysis is provided in Section 4.2.2.

4. Case study

This section presents a case study of the travel behavior of HongKong inbound tourists using geotagged photos. It starts with adescription of our data collection, followed by the analysis andresults. We then set out the practical implications of the findings tosupport Hong Kong tourism managers in improving theirperformance.

4.1. Data collection

The dataset used in this paper is collected from Flickr followingthe method described in Section 3.1. Since our focus is on HongKong inbound travelers, we firstly search for seed photos byproviding a bounding box for the Hong Kong area in the photosearch function, as shown in Table 1.

The coordinates of the bounding box are in a decimal degreeform, which can be determined manually by using Google Maps(maps.google.com). We select these values to ensure that thebounding box covers the geographical area of Hong Kong entirely.We only retrieve photos taken from 2011 to August 28, 2013 inorder to ensure the dataset is up to date. Accordingly, only theearliest photo taken date parameter (tmin) is provided.

From the seed photos, a list of owner identification numbers isextracted. Those indicating the owner's origin is not from HongKong are treated as labeling inbound travelers and hence used toretrieve their entire photo collection. The metadata tags of eachphoto contain photoID, OwnerID, owner location of origin, date andtime taken, and GPS location (longitude, latitude). The GPS locationindicates where the photo was taken, which reflects the tourist'stravel footprint. If a user takes many photos at the same location,we keep only one of those and discard the others.

Since people from different countries tend to have differenttravel preferences (Leung et al., 2012), we group tourists based ontheir location of origin to examine the behavior of travelers withdifferent profiles. We observe that the majority of tourists visitingHong Kong are from countries in the Asia-Pacific, European, andNorth American regions. The European and North American tour-ists are mainly from English-speaking countries such as the U.S.,Canada, and the United Kingdom. We can assume that they aremore similar to one another in terms of background than touristsfrom Asia. Therefore, we group tourists from Europe and NorthAmerica together and refer to them henceforward as Westerntourists, and similarly group tourists from Asia together and labelthis group as Asian tourists. We arrive at a dataset containing29,443 photos collected from 2100 Hong Kong inbound tourists asshown in Table 2.

The following sections demonstrate how the spatial and tem-poral information in geotagged photos can help to address thechallenges of travel analysis. More specifically, we perform threemajor analyses to examine the behavior of inbound tourists in Hong

Table 1Parameters for photo search function.

Parameter Value Description

xmin 113.887603 Minimum longitude of the bounding boxymin 22.215377 Minimum latitude of the bounding boxxmax 114.360015 Maximum longitude of the bounding boxymax 22.51446 Maximum latitude of the bounding boxtmin 1/1/2011 Earliest photo taken datetmax e Latest photo taken date

Kong, namely AOI Identification, Tourist Movement Analysis, andTourist Activity Analysis.

4.2. Findings and analysis

4.2.1. AOI IdentificationThis section presents our findings on the identification of the

AOI for Hong Kong inbound travelers to support managers indeveloping their tourism locations. Firstly, we inspect our datacollection by visualizing the photo locations by incorporating theirGPS information into Google Earth software, as shown in Fig.1. Eachphoto is indicated by a white dot on the satellite image.

The geotagged photos appear all over the regions of Hong Kong,which indicate that the raw geographical information is noisy andmay be misleading if directly used for analysis. We therefore applythe P-DBSCAN clustering algorithm to remove the noise and iden-tify the locations in which tourists are most interested. An advan-tage of this density clustering approach is that users are notrequired to provide the number of clusters in the algorithm; theclusters are automatically determined based on the density of thepoints and number of visitors. We set the neighborhood radiusvalue to d¼ 0.002, which is equivalent to approximately 150 m. Theminimum owner (l) is set to a value of 10% of the total number oftourists.

A total of seven clusters are found, indicating there are sevenareas of interest for Hong Kong inbound tourists, as shown in Fig. 2.Four of them are in the Hong Kong metropolitan area, includingCenter Mong Kok, the Tsim Sha Tsui area, Hong Kong Central, andTimes Square Towers. Two areas of interest are found in thecountryside, namely the Peak Tower and the Tian Tan BuddhaStatue. Hong Kong International Airport is also an AOI, as indicatedby the many photos taken within its terminals. These findings areconsistent with the fact that the metropolitan area of Hong Kong iswell known as a major destination for visitors. P-DBSCAN is shownto be effective in removing noise from geotagged photos andidentifying AOI to inbound tourists.

In order to identify the location preferences of different groupsof tourists, we also take their profile into account. The clusters areexamined with respect to the two groups, Asian and Westerntourists, and the proportion of holiday makers visiting each area isused as a measure of its popularity, as shown in Table 3. The AOI ispresented in a ranking order from the most to the least popular.

� Hong Kong Central and the Tsim Sha Tsui area are the two mostpopular destinations for tourists in both groups, as shown bytheir ranking as first and second. Hong Kong Central is thecentral business district and the “heart” of Hong Kong, while theTsim Sha Tsui area is a major tourist hub in the metropolitanarea with many shops, restaurants, and a good view of VictoriaHarbor.

� Western tourists show more interest in the Peak Tower (third)and Center Mong Kok (fourth) than in Times Square Towers(fifth) and Hong Kong International Airport (sixth). Asian tour-ists show the opposite trend, as indicated by their ranking ofTimes Square Towers and Hong Kong International Airport asmore popular than the Peak Tower and Center Mong Kok.

Fig. 1. Photo locations of Hong Kong inbound tourists.

Fig. 2. Areas of interest for inbound tourists.

Table 3Popularity of areas of interest.

Group Area of interest Percentage(%)

Popularityrank

Asian tourist Hong Kong Central 40.35 1Tsim Sha Tsui Area 38.80 2Times Square Towers 20.08 3Hong Kong InternationalAirport

18.34 4

The Peak Tower 14.19 5Center Mong Kok 10.52 6

Westerntourist

Hong Kong Central 47.92 1Tsim Sha Tsui Area 44.64 2The Peak Tower 19.74 3Center Mong Kok 15.14 4Times Square Towers 12.22 5Hong Kong InternationalAirport

10.99 6

Tian Tan Buddha Statue 10.43 7

H.Q. Vu et al. / Tourism Management 46 (2015) 222e232 227

� It is interesting to see that the Tian Tan Buddha Statue is iden-tified as an AOI for Western but not Asian tourists. A possibleexplanation is that most Asian countries are likely to havesomewhat similar cultural or religious sites such as temples,shrines, or Buddha statues. Asian visitors may, therefore, not beinterested in seeing similar things when visiting Hong Kong. Onthe other hand, Western people are likely to be interested inexploring Asian culture, and the Tian Tan Buddha Statue is onesuch symbol.

4.2.2. Tourist Movement AnalysisKnowledge of tourist movements is important in transportation

planning and traffic management, especially for metropolitan areaswith a high density of traffic. An overview of how tourists movefrom one area to another andwhat routes they prefer to takewill bebeneficial from developing appropriate travel management plans.To address such challenges, we therefore focus this part of theanalysis on tourist movements between the AOI in the metropol-itan district. To represent travel behavior, we construct the travel

H.Q. Vu et al. / Tourism Management 46 (2015) 222e232228

trajectories of tourists corresponding to spatial and temporalsequence. The GPS information from the photos is concatenatedaccording to the time taken on a daily basis. Fig. 3 represents themovement trajectory of tourists as generated from their geotaggedphotos between five areas of interest in Hong Kong city center. Eachtrajectory is displayed as a thin white line. Regions and paths withmany lines passing through them indicate a high density ofmovement. It appears that tourists tend to travel between areasthat are close to each other, such as from Center Mong Kok to TsimSha Tsui, Tsim Sha Tsui to Hong Kong Central, or Hong Kong Centralto Times Square Towers.

To gain more detailed knowledge of the travel flow and di-rections of different tourist groups, we analyze tourist trajectoriesbetween AOI by applying the Markov Chain technique. A one-steptransition probability matrix is computed for each travel group asdescribed in Section 3.3. A high value for P(ajjai) suggests thattourists are likely to visit aj right after ai. To make interpretationeasier, we display in Fig. 4 the transition probability and transitiondirection for Asian and Western tourists. Only P(ajjai) � 0.3 isshown. Our findings can be summarized as follows:

� Fig. 4a shows that Asian tourists are likely to flow to Hong KongCentral from the surrounding areas, as shown by the dark ar-rows and a high probability value of around 0.6. Tourists inCenter Mong Kok tend to visit Tsim Sha Tsui next (dark arrowwith a probability of above 0.5), while some chose to travel ondirectly to Hong Kong Central. Considerable tourist flow is foundfrom Hong Kong Central and the Times Square Towers to TsimSha Tsui with probabilities of 0.44 and 0.307, respectively.

� Fig. 4b shows some differences between Western and Asiantourists. Namely, Tourists from Asian countries tend to visit

Fig. 3. Movement trajectory of tourist

Hong Kong Central right after Center Mong Kok (dark arrowwith a probability of 0.678), while some travel in the oppositedirection. Western tourists are more likely to visit Tsim Sha Tsui(probability of 0.579) than Hong Kong Central (probability of0.316) after Times Square Towers. Some tourists visited Tsim ShaTsui immediately after the Peak Tower.

We further perform z-tests to verify the statistical significancefor the common flows between Asian and Western tourists at p-value < 0.05, as shown in Table 4. Based on Table 4 and Fig. 4, theflows of Western tourists from Center Mong Kok to Hong KongCentral and from Times Square Towers to Tsim Sha Tsui aresignificantly higher than the flows of Asian tourists (flows F3 andF6). However, Asian tourists are more likely to flow from TimesSquare Towers to Hong Kong Central than Western tourists (flowF4). The other flows of Asian and Western tourists are not signifi-cantly different from each other as shown by p-value > 0.05 (flowsF1, F2 and F5).

Although Fig. 4 indicates the travel flow of tourists from onelocation to another, it is not necessarily the actual routes whichthey take. Such routes can be revealed by analyzing the photostakenwhile traveling between AOI. To demonstrate this, we inspectthe common travel routes taken by tourists between more widelyseparated areas such as Center Mong Kok and Times Square Towers.Firstly, we identify tourists who visited both locations on a singleday. Then, we retrieve and plot the photos taken while travelingfrom one place to another in Fig. 5. The white dots denote photoswhich represent the footprint of tourists during their trip. It can beseen that most tourists traveled to and from these locations viaTsim Sha Tsui and Hong Kong Central. To be more specific, theytravel along Nathan Road between Center Mong Kok and Tsim Sha

generated from geotagged photos.

Fig. 4. Tourist traffic flow in Hong Kong metropolitan area.

H.Q. Vu et al. / Tourism Management 46 (2015) 222e232 229

Tsui. Direct ferry services are used to travel between Tsim Sha Tsuiand Hong Kong Central. To go from Hong Kong Central and TimesSquare Towers tourists use Hennessy Road. We also note an inter-esting finding about tourist preferences. There is a direct ferryservice which departs from Wan Chai, which is near Times SquareTowers and goes to Tsim Sha Tsui, but most tourists did not choosethis option and preferred to travel to Tsim Sha Tsui via the ferry lineleaving from Hong Kong Central.

4.2.3. Tourist Activity AnalysisTourism managers are not only interested in knowing where

and how tourists travel, but also when they visit places. This iscrucial in transportation planning and destination management, toavoid problems of overload when too many people visit the sameplace in a short time frame. To address this, we also examine thevisiting pattern of tourists based on the spatial and temporal in-formation in their photos. More specifically, we compute theprobability of tourists appearing at particular AOI over a 24-hperiod, as shown in Fig. 6. The horizontal axis represents time, andthe vertical axis represents probability. Here, only six AOI commonto both Asian and Western tourists are included, because these arethe locations which receive many visits. We summarize the find-ings as follows:

� Fig. 6a shows the presence of tourists at Hong Kong Interna-tional Airport on a given day. Asian tourists appear to be arrivingor departing during the daytime (10:00e17:00), whereasWesterners may travel at any time.

� Fig. 6b shows that tourists in both groups are likely to visit HongKong Central between 11:00 and 16:00. In particular, the peaktime for Asians is at noon, while the busiest time for Westernersis 15:00.

Table 4Z-test result on travel flow between Asian and Western tourists. (Significant level p-value ¼ 0.05).

Tourist traffic flow Z-score p-Value ID

Hong Kong Central / Tsim Sha Tsui �1.000 0.317 F1Tsim Sha Tsui Area / Hong Kong Central �0.162 0.873 F2Center Mong Kok / Hong Kong Central �3.490 0.000 F3Times Square Towers / Hong Kong Central 3.952 0.000 F4The Peak Tower / Hong Kong Central 1.636 0.101 F5Times Square Towers / Tsim Sha Tsui �3.507 0.000 F6

� Tourists are likely to visit Tsim Sha Tsui area in the late afternoonand evening, as shown in Fig. 6c. Asian tourists aremost likely tobe there at 17:00 and 21:00 h, While Western tourists visit mostoften at around 20:00. Similarly, the busy time for Times SquareTowers is in the afternoon, as shown in Fig. 6d.

� Similar visiting patterns to Center Mong Kok are found for bothAsian and Western visitors, as shown in Fig. 6e. Both groupsdemonstrate a peak visiting time at this location of 16:00. Theyalso have relatively similar visiting patterns for the Peak Toweras shown in Fig. 6f, where the peak for Asian tourists is 19:00.

4.3. Discussion

The analysis of AOI in Section 4.2.1 highlights some differencesin travel preferences for different tourist groups. Different travelpackages and tour routes can be developed accordingly to meet the

Fig. 5. Travel route of tourist between Center Mong Kok and Times Square Towers.

Fig. 6. Tourist activities in areas of interest.

H.Q. Vu et al. / Tourism Management 46 (2015) 222e232230

needs of inbound tourists. For instances, Western, but not Asian,tourists show an interest in Tian Tan Buddha Statue. Tourismmanagers can thus develop marketing plans and transportationarrangement to promote this attraction to Western visitors. Thepopularity ranking given in Table 3 can also help tourism de-velopers to design local tour recommendations for tourists. Tours toTimes Square Towers can be recommended to Asian visitors beforethe Peak Tower and Center Mong Kok, whereas tours to the PeakTower and CenterMong Kok can be recommended toWesterners inpreference to Times Square Towers.

The analysis of tourist movement in Section 4.2.2 can also offersome practical implication for traffic management. For example,the traffic flow captured in Fig. 4 suggests that additional trans-portation to Hong Kong Central could be arranged for Asian visitors,as they are likely to visit this location from surrounding areas. Onthe other hand, more direct transportation means can be arrangedfor Western tourists to travel from Center Mong Kok to Hong KongCentral, or from Times Square Towers to Tsim Sha Tsui area. Inaddition, Fig. 5 shows the current travel routes taken by touristsbetween Center Mong Kok and Times Square Towers. Most peopletend to use the ferry service from Hong Kong Central to Tsim ShaTsui, although there is actually a direct ferry service leaving from

Wan Chai, which is near the Times Square Towers. Traffic managerscan thus develop appropriate management plans to encouragetourists to use the direct service. This can reduce the traffic inCauseway Bay and the ferry line between Hong Kong Central andTsim Sha Tsui, helping to avoid traffic congestion in the future asmore visitors will travel to Hong Kong.

The tourist activity analysis presented in Section 4.2.3 reflectsthe visiting patterns of tourists to a particular AOI. The knowledgeof where tourists are likely to be at different times of the day cansupport destination promotion and management. For instance,supporting service and transportation can be arranged at the HongKong International Airport, where many tourists are arriving anddeparting (Fig. 6a). Different travel itineraries can be developed toavoid overcrowding in Hong Kong Central around noon (Fig. 6b), orat Center Mong Kok in the afternoon (Fig. 6e).

5. Conclusions

Knowledge of travel behavior is crucial to help tourism man-agers construct strategic plans andmake decisions that will create asustainable tourism industry. In spite of existing research efforts,managers still face significant challenges in gaining insight into

H.Q. Vu et al. / Tourism Management 46 (2015) 222e232 231

tourist travel behavior due to the limitations of existing approacheson data collection such as surveys and polls. Such methods areunable to capture travel behavior comprehensively, nor to accu-rately reflect travel patterns. Travel behaviors vary among touristswith different profiles and so far only limited attempts have beenmade to take the preferences of different groups into consideration.An efficient method of comprehensively capturing the travelbehavior patterns of tourists is therefore required.

To address these challenges, we approached the task of mappingtourist travel behavior by exploiting the socially generated anduser-contributed geotagged photos that are publicly available onthe Internet. We presented a method for constructing a traveldataset from geotagged photos on Flickr, one of the most popularwebsites for sharing photos. A dataset containing thousands ofphotos with temporal and geographic information attached enablesus to capture themovement trajectories of tourists on a larger scale.In addition, we introduced two techniques, P-DBSCAN and theMarkov Chain, to mine travel behavior patterns from this dataset.We demonstrated the effectiveness of these approaches by usingthe case study of Hong Kong inbound tourism and discovered thelocations of interest to travelers, their travel patterns, and theirdaily activities. Practical implications are offered to support HongKong tourism managers in destination development, trans-portation planning, and impact management.

Since the collection of data from geotagged photo sets is notlimited by factors of time and space, our future work will focus onanalyzing tourist travel behavior on a larger scale (international orglobal). This will enable data from larger numbers of tourists acrossmany years to be considered in order to model changes and trendsin travel behavior for different tourism markets. Besides the spatialand temporal information, the online travel photos can provide thetextual and visual information. The information would have greatpotential for providing insight into travelers' behavior, which canthen be further explored in further research. Moreover, the analysiscan be carried out with more detailed information about touristprofiles such as gender and age. Depending on practical applica-tions, future work can reveal deeper insight into tourist behaviors.

Acknowledgment

The work described in this paper was supported by a grantfunded by the Research Grants Council of the Hong Kong SpecialAdministrative Region, China (GRF Project Number: 15503814).This project was also funded by the Hong Kong Polytechnic Uni-versity (G-UC86), the Hong Kong Scholars Program (G-YZ28), andthe National Natural Science Foundation of China (71361007).

References

Asakura, Y., & Iryo, T. (2007). Analysis of tourist behaviour based on the trackingdata collected using a mobile communication instrument. TransportationResearch Part A, 41(7), 684e690.

Ashley, C., Brine, P. D., Lehr, A., & Wilde, H. (2007). The role of the tourism sector inexpanding economic opportunity. Corporate Social Responsibility Initiativereport 23. Cambridge, MA: Kennedy School of Government, Harvard University.

Batra, A. (2009). Senior pleasure tourists: examination of their demography, travelexperience, and travel behavior upon visiting the Bangkok metropolis. Inter-national Journal of Hospitality & Tourism Administration, 10(3), 197e212.

Chancellor, H. C. (2012). Applying travel pattern data to destination developmentand marketing decisions. Tourism Planning and Development, 9(3), 321e332.

Deng, M., & Athanasopoulos, G. (2011). Modelling Australian domestic and inter-national inbound travel: a spatialetemporal approach. Tourism Management,32(5), 1075e1084.

Edwards, D., Griffin, T., Hayllar, B., Dickson, T., & Schweinsberg, S. (2009). Under-standing tourist experience and behaviour in cities: An Australian case study.Technical report. Sustainable Tourism. Available online at http://www.sustainabletourismonline.com/31/destination-access/understanding-tourist-experiences-and-behaviour-in-cities-an-australian-case-study.

Ester, M., Kriegel, H. P., Sander, J., & Xu, X. (1996). A density based algorithm fordiscovering clusters in large spatial database with noise. In Proceedings of 2nd

international conference on knowledge discovery and data mining (pp. 226e231).Portland, Oregon, USA.

Hays, J., & Efros, A. A. (2008). Im2gps: estimating geographic information from asingle image. In Proceedings of the IEEE conference on computer vision andpattern recognition (pp. 1e8). Anchorage, Alaska, USA.

Hwang, Y. H., Gretzel, U., & Fesenmaier, D. R. (2006). Multi-city trip patterns touriststo the United States. Annals of Tourism Research, 33(4), 1057e1078.

Jaffe, A., Naaman, M., Tassa, T., & Davis, M. (2006). Generating summaries andvisualization for large collections of geo-referenced photograph. In Proceedingsof the 8th ACM international workshop on multimedia information retrieval (pp.89e98). Santa Barbara, California, USA.

Kalogerakis, E., Vesselova, O., Hays, J., Efros, A. A., & Hertzmann, A. (2009). Imagesequence geolocation with human travel priors. In Proceedings of the IEEE in-ternational conference on computer vision (pp. 253e260). Tokyo, Japan.

Kennedy, L., Naaman, M., Ahern, S., Nair, R., & Rattenbury, T. (2007). How Flickrhelps us make sense of the world: context and content in community-contributed media collections. In Proceedings of the 15th international confer-ence on multimedia (pp. 631e640). Augsburg, Germany.

Kisilevich, S., Keim, D., & Rokach, L. (2010). A novel approach to mining travel se-quences using collections of geotagged photos. In Geospatial thinking (pp.163e182). Berlin; Heidelberg: Springer.

Kisilevich, S., Mansmann, F., & Keim, D. (2010). P-DBSCAN: a density based clus-tering algorithm for exploration and analysis of attractive areas using collec-tions of geo-tagged photos. In Proceedings of the 1st international conference andexhibition on computing for geospatial research & application. Article no. 38,Bethesda, Maryland, USA.

Kurashima, T., Iwata, T., Irie, G., & Fujimura, K. (2010). Travel route recommendationusing geotags in photo sharing sites. In Proceedings of the 19th ACM internationalconference on information and knowledge management (pp. 579e588). Toronto,Canada.

Lau, G., & McKercher, B. (November 2006). Understanding tourist movement pat-terns in a destination: a GIS approach. Tourism and Hospitality Research, 7,39e49.

Lee, I., Cai, G., & Lee, K. (2013). Mining points-of-interest association rules from geo-tagged photos. In Proceeding of the 46th Hawaii international conference onsystem sciences (pp. 1580e1588). Grand Wailea, Maui, Hawaii, USA.

Leung, X. Y., Wang, F., Wu, B., Bai, B., Stahura, K. A., & Xie, Z. (2012). A social networkanalysis of overseas tourist movement patterns in Beijing: the impact of theOlympic Games. International Journal of Tourism Research, 14, 469e484.

Lew, A., & McKercher, B. (2006). Modeling tourist movement a local destinationanalysis. Annals of Tourism Research, 33(2), 403e423.

Li, X. (2013). Multi-day and multi-stay travel planning using geotagged photos. InProceeding of the 2nd ACM SIGSPATIAL international workshop on crowdsourcedand volunteered geographic information (pp. 1e8). Orlando, Florida, USA.

Li, X., Meng, F., & Uysal, M. (2008). Spatial pattern of tourist flows among the Asia-Pacific countries: an examination over a decade. Asia Pacific Journal of TourismResearch, 13(3), 229e243.

Mamei, M., Rosi, A., & Zambonelli, F. (2010). Automatic analysis of geotagged photosfor intelligent tourist services. In Proceeding of the 6th international conferenceon intelligent environments (pp. 146e151). Kuala Lumpur, Malaysia.

Majid, A., Chen, L., Chen, G., Mirza, H. T., & Hussain, I. (2012). Spatial pattern oftourist flows among the Asia-Pacific countries: an examination over a decade.Asia Pacific Journal of Tourism Research, 13(3), 229e243.

Majid, A., Chen, L., Chen, G., Mirza, H. T., Hussain, I., & Woodward, J. (2013).A context-aware personalized travel recommendation system based on geo-tagged social media data mining. International Journal of Geographical Infor-mation Science, 27, 662e684.

Masiero, L., & Zoltan, J. (2013). Tourist intra-destination visits and transport mode: abivariate probit model. Annals of Tourism Research, 43(4), 529e546.

McKercher, B., & Lau, G. (2008). Movement patterns of tourists within a destination.Tourism Geographies: An International Journal of Tourism Space, Place and Envi-ronment, 10(3), 355e374.

McKercher, B., Shoval, N., Ng, E., & Birenboim, A. (2011). First and repeat visitorbehaviour: GPS tracking and GIS analysis in Hong Kong. Tourism Geographies: AnInternational Journal of Tourism Space, Place and Environment, 14(1), 147e161.

Okuyama, K., & Yanai, K. (2013). A travel planning system based on travel trajectoriesextracted from a large number of geotagged photos on the web. The era of inter-active media (pp. 657e670). Springer New York.

Orellana, D., Bregt, A. K., Ligtenberg, A., & Wachowicz, M. (2012). Exploring visitormovement patterns in natural recreational areas. Tourism Management, 33,672e682.

Prideaux, B. (2000). The role of the transport system in destination development.Tourism Management, 21(1), 53e63.

Quack, T., Leibe, B., & Gool, L. V. (2008). World-scale mining of objects and eventsfrom community photo collections. In Proceedings of the 7th international con-ference on content-based image and video retrieval (pp. 47e56). Niagara Falls,Canada.

Rugna, J., Chareyron, G., & Branchet, B. (2012). Tourist behavior analysis throughgeotagged photographies: a method to identify the country of origin. In Pro-ceedings of the 13th IEEE international symposium on computational intelligenceand informatics (pp. 347e351). Budapest, Hungary.

Shi, Y., Serdyukov, P., Hanjalic, A., & Larson, M. (2013). Nontrivial landmarkrecommendation using geotagged photos. ACM Transaction on Intelligent SystemTechnology, 4(3). Article no. 47.

H.Q. Vu et al. / Tourism Management 46 (2015) 222e232232

Smallwood, C. B., Beckley, L. E., & Moore, S. A. (2012). An analysis of visitor move-ment patterns using travel networks in a large marine park, north-westernAustralia. Tourism Management, 33(3), 517e528.

Snavely, N., Seitz, S. M., & Szeliski, R. (2008). Modeling the world from internetphoto collections. International Journal of Computer Vision, 80(2), 189e210.

Wu, C. L., & Carson, D. (2008). Spatial and temporal tourist dispersal analysis inmultiple destination travel. Journal of Travel Research, 46(3), 311e317.

Xia, J., Evans, F., Spilsbury, K., Ciesielski, V., Arrowsmith, C., & Wright, G. (2010).Market segments based on the dominant movement patterns of tourist. TourismManagement, 31(4), 464e469.

Xia, J., Zeephongsekul, P., & Arrowsmith, C. (2009). Modelling spatio-temporalmovement of tourists using finite Markov chains. Mathematics and Computersin Simulation, 79(5), 1544e1553.

Yanai, K., Kawakubo, H., & Qiu, B. (2009). A visual analysis of the relationship be-tween word concepts and geographical locations. In Proceedings of the ACMinternational conference on image and video retrieval. Article no. 13, SantoriniIsland, Greece.

Yanai, K., Yaegashi, K., & Qiu, B. (2009). Detecting cultural differences usingconsumer-generated geotagged photos. In Proceedings of the 2nd internationalworkshop on location and the web. Article no. 12, Boston, Massachusetts, USA.

Yin, H., Wang, C., Yu, N., & Zhang, L. (2012). Trip mining and recommendation fromgeo-tagged photos. In Proceedings of the IEEE international conference onmultimedia and expo workshops (pp. 540e545). Melbourne, Australia.

Zakrisson, I., & Zillinger, M. (2012). Emotions in motion: tourist experiences in timeand space. Current Issues in Tourism, 15(6), 505e523.

Zheng, Y. T., Zha, Z. J., & Chua, T. S. (2011). Research and applications on georefer-enced multimedia: a survey. Multimedia Tools and Applications, 51(1), 77e98.

Zheng, Y. T., Zha, Z. J., & Chua, T. S. (2012). Mining travel patterns from geotaggedphotos. ACM Transactions on Intelligent Systems and Technology, 3(3), 1e18.

Huy Quan Vu is a Ph.D. student at the School of Infor-mation Technology, Deakin University. His research in-terests include machine learning and data miningapplications in tourism and hospitality.

Gang Li, Ph.D., is a Senior Lecturer at the School of Infor-mation Technology, Deakin University. His research in-terests are machine learning, data mining, and technologyapplications to tourism and hospitality.

Rob Law, Ph.D., is a Professor at the School of Hotel andTourism Management, the Hong Kong Polytechnic Uni-versity. His research interests are information manage-ment and technology applications.

Ben Ye, Ph.D., is an Assistant Professor at School of Busi-ness, Sun Yat-sen University. His research interests includeconsumer behavior and cross-cultural consumerpsychology.